Comparable Set: The Group of Companies That Carries the Range

Definition

The comparable set is the study’s final group of comparable companies — the companies that survived the quantitative screening and the qualitative screening, whose profit level indicator values are pooled to build the arm’s length range. It is the input population of the statistics: the median, the percentiles, and the interquartile range are computed on this set — so the set’s composition is the study’s most consequential design decision, above the method and the PLI (the method and PLI say what to measure; the set says on whom). The set is defined by three properties: size (enough members for the statistics to be meaningful — the statistical significance question), consistency (every member comparable to the tested party on the five comparability factors, on the same PLI, on the same data periods, with adjustments applied uniformly), and documented composition (the [accept-reject matrix] (/docs/glossary/accept-reject-matrix) that records why each member is in and each non-member is out).

The comparable set, in one build:
  1. The search population (the candidates the search string returned)
  2. The screened candidates (the quantitative filters applied)
  3. The reviewed candidates (the company-by-company qualitative check)
  4. The set (the accepts — the members, the values, the adjustments)
  5. The record (the matrix — the reasons, the sources)

The property	The standard
Size	Enough members for the distribution to carry a range — the practice floor is discussed in the statistical significance term (fewer than ~4 values, the quartiles are near-degenerate; the TPO’s recurring challenge to small sets)
Consistency	Same PLI, same periods (the multi-year averaging rule applied uniformly), same adjustment set (the working capital adjustment and any other, applied to every member or none), same segment basis
Composition record	The accept-reject matrix — every screened candidate accounted for, with the specific reason and source for the decision
Stability	The set is fixed for the study — mid-study additions or removals (a “one more comparable” after seeing the range) are the pattern the examination reads as fitting the range to the result

The working read (the benchmarking study guide): the set is where the methodology stops being a procedure and becomes a claim — the claim that these companies’ returns are an arm’s length proxy for the tested party’s. The TPO’s standard sequence of attacks runs: (1) the set is too small (the significance objection), (2) a member is not comparable (the matrix check — the reason column must hold), (3) a comparable was excluded that should have been included (the other side of the matrix), and (4) the set was chosen to produce the range (the stability objection — the set should be the output of the screens, not the input of the preference). A set that is sized, consistent, and recorded with specific reasons survives all four; a set that is thin, mixed, or unreasoned fails at least one.

Example

A TNMM study (OP/OC PLI, 3-year simple average) on an Indian contract manufacturer: the search (NIC 312/319, the manufacturing keywords) returns 58 candidates; the quantitative screens (the size floor, the profitability filter, the asset-intensity filter) reduce it to 22; the qualitative review (product type, capacity utilisation, the toll-vs-contract distinction) decides: 14 accepted, 7 rejected (the recorded reasons: 3 × different product class, 2 × loss-making with structural causes, 1 × toll-only model, 1 × segment data unavailable), 1 flagged then excluded (extraordinary item that could not be reliably normalized). The comparable set is the 14: same PLI (OP/OC), same 3 years, working-capital adjusted uniformly, the IQR computed on the 14 values. The Local File’s benchmarking annex carries the matrix (all 22 decisions) and the set table (the 14 values, the adjusted values) — and the set’s composition is what the TPO’s counsel tests, not the arithmetic.

See also

• Comparable Companies
• Accept-Reject Matrix
• Statistical Significance
• Benchmarking Study Guide

FAQ

How many comparables does a set need? There is no statutory number; the practical floor comes from the statistics: below about 4 values the quartiles are near-degenerate (a 4-value IQR is two middle points, the range is a segment between two companies’ margins), and the statistical significance of the result is unprovable. Practice targets 10–25 members where the search space allows; a set of 3 is defensible only with an explicit justification of why the market is thin (the niche product, the young industry) — the justification being part of the record, not an afterthought.

Can the set differ from study to study for the same tested party? It should — the set is built from the current search, the current data, the current screening; a refreshed study (the benchmark refresh) legitimately produces a different composition. What it must not do is drift within a year or a defence without a recorded reason: the set that changes when the tested party’s result moves outside the range is the stability objection in action. Same tested party, same year, different set → the matrix must show why.

Is the comparable set the same as the “comparables” shown in the report? The report’s comparables table is the set’s presentation — the accepted members with their values. The set is the statistical population; the table is its display. In the file, the set is also the decision record (the matrix around it). Examination practice: the reviewer works from the table but audits the set — the members, the exclusions, the adjustments — because the range inherits every quality (and every defect) of the set that produced it.